This paper compares different large language models (LLMs) on the task of detecting violations of cease and desist declarations in German online product descriptions as part of the KIVEDU project. We evaluate two proprietary LLMs by OpenAI (gpt-3.5-turbo-0301 and gpt-3.5-turbo-0613) and three open source LLMs by various organizations (LLaMA2, StableBeluga2, and Platypus2) using different variations in text input (prompt) on a dataset of 116 manually labeled pairs of cease and desist declarations and product descriptions. The evaluation aims to explore two research questions: 1) Which LLM is most adept at identifying violations? and 2) How do prompt variations impact model performance? The results show that StableBeluga2 performed best, achieving the highest accuracy and micro F1 score. It was also the most reliable model with minimal deviations in performance across prompt variants. The Platypus2 and gpt-3.5-turbo-0301 models also achieved good results though they displayed greater variability in their performance. The worst-performing model was LLaMA2. The results further show that prompting had a significant impact on model performance, with the presence of a step-by-step instruction generally decreasing performance and a “yes”/“no” output format leading to higher performance. However, this was highly dependent on the specific model. Role prompting and providing a longer vs. shorter instruction had a minimal impact on performance across models. Overall, the study demonstrates the potential of LLMs in automating the detection of cease and desist violations in online product descriptions. Further research is needed to evaluate other LLM models and prompt variations, as well as to explore approaches like LLM fine-tuning on domain-specific data to further improve performance.